We study stochastically perturbed non-holonomic systems from a geometric point of view. In this setting, it turns out that the probabilistic properties of the perturbed system are intimately linked to the geometry of the constraint distribution. For G-Chaplygin systems, this yields a stochastic criterion for the existence of a smooth preserved measure. As an application of our results, we consider the motion planning problem for the noisy two-wheeled robot and the noisy snakeboard.